Partition
Divide the 1,024-token shape sequence into contiguous blocks.
Video
Block3D in motion
Abstract
Fast generation without giving up correction
While text-to-3D generation has advanced rapidly, achieving high geometric fidelity at low inference cost remains challenging. Existing methods either decode discrete shape tokens autoregressively or iteratively refine global 3D representations with diffusion or flow models. Autoregressive decoding is sequential and cannot revise errors, whereas diffusion and flow-matching models repeatedly process the full representation.
We propose Block3D, a block-wise diffusion framework that partitions the discrete shape-token sequence into contiguous blocks, generates the blocks autoregressively, and jointly denoises all tokens within the current block. Confidence-guided intra-block correction revises low-confidence tokens before each block is finalized. On a held-out set from TRELLIS-500K, Block3D reduces mean end-to-end generation time from 25.71 seconds to 4.99 seconds, achieving a 5.15× speedup over the fine-tuned autoregressive baseline without sacrificing geometric fidelity.
5.15×faster than AR
4.99 send-to-end generation
0.309F-score at 1%
0.668normal consistency
Interactive 3D Results
Generated meshes, up close
Nine higher-detail Block3D generations, rendered directly from the generated geometry.
Comparisons
Quality across prompts
Front and back views use the same camera and normalization for every method.
A stylized knight character with a hexagonal helmet with a horizontal eye slit, rigid armor plates, a rectangular shield, and a sword arranged in a combat-ready pose.
Method
Block-wise diffusion
Block3D retains causal structure across blocks while enabling parallel, bidirectional denoising within each active block.
Denoise
Update every token in the active block jointly and in parallel.
Correct
Revise low-confidence tokens before committing the current block.
Decode
Pass the completed sequence through the frozen shape decoder.
More Results
From assets to a scene
Independently generated components are composed with shared lighting and cameras.
Citation
Cite Block3D
Please cite the arXiv preprint using the BibTeX entry.
@article{block3d2026,
title = {Block3D: Efficient Text-to-3D Generation
via Block-Wise Diffusion},
author = {Cui, Bowen and Wang, Weijie and Zhang, Zeyu and
He, Yefei and Lin, Mingda and Zhao, Haoyu and
He, Yuanyu and Chen, Donny Y. and Chen, Feng and
Zhuang, Bohan},
journal = {arXiv preprint},
year = {2026}
}